1. Field of the Invention
The present invention relates to image processing, and more particularly to a technique capable of detecting an imaging object pattern from a moving image.
2. Description of the Related Art
Conventionally, an image processing method which can automatically detect a predetermined imaging object pattern from a moving image has been discussed and the method can be used to determine, for example, a human face or the like. This kind of image processing method can be used in a video conference, a monitoring system, or the like.
Such techniques that can be used to detect a predetermined imaging object pattern from an image are, for example, discussed in Japanese Patent Application Laid-Open No. 2007-25899, Japanese Patent Application Laid-Open No. 2004-171490, and Japanese Patent Application Laid-Open No. 2003-235035 according to which the predetermined imaging object pattern can be detected using a template matching technique.
However, according to the technique discussed in Japanese Patent Application Laid-Open No. 2007-25899, a burden in calculation processing increases if the system generates a plurality of reduced images from an input image on a frame-by-frame basis and performs matching between a predetermined imaging object pattern and the template. Accordingly, unless an employed hardware has high-speed processing capabilities, it will be required to use the method discussed in Japanese Patent Application Laid-Open No. 2004-171490 to perform detection processing according to the template matching at constant frame intervals so that the calculations for processing can be smoothly executed.
In this case, for example, if a zoomed-up image of a predetermined imaging object is captured by a camera, the system may not perform tracking using panning and tilting functions when the predetermined imaging object exits from the frame between non-processed frames.
Further, according to the technique discussed in Japanese Patent Application Laid-Open No. 2004-171490, the system performs normal detection processing at constant frame intervals and inserts a frame that detects only a position adjacent to a coordinate of the already detected imaging object between two frames at which the normal detection processing is performed. This system intends to reduce the burden in calculations for the template matching processing.
However, the imaging object may not be captured in a detection area if a moving speed of the imaging object is inappropriate. The imaging object cannot be detected in this case. Further, in a zoom-up operation of the camera, or in a situation where a distance between the imaging object and the camera becomes shorter, the imaging object may not be detected based on only the detection at or near the coordinate position of the previous detection frame.
Further, according to the technique discussed in Japanese Patent Application Laid-Open No. 2003-235035, the system obtains a difference between an input image and a background image and registers the obtained difference as a template. In this case, a predetermined imaging object in the template is identical to a predetermined imaging object in the original input image.
If the camera performs a zooming control, the system performs conversion in resolution based on the zooming parameter to equalize in size the predetermined imaging object of the input image with the predetermined imaging object of the template, so that the imaging object pattern can be detected.
However, this technique is only effective in an operation for tracking an imaging object that has been captured in the input image and cannot be used in a situation where the distance between the imaging object and the camera is variable. Accordingly, detection accuracy deteriorates when the distance between the imaging object and the camera is changed.